Learning method and learning device for generating virtual feature map having the same or similar characteristics as real feature map using Gan capable of being applied to domain adaptation used in the virtual travel environment, and test method and test device using it

摘要

To alleviate a problem in which sets of training images in a non-RGB format are produced by transforming sets of training images in an RGB format into those in the non-RGB format through a cycle generative adversarial network (GAN) capable of being applied to domain adaptation.SOLUTION: A learning method for deriving virtual feature maps from virtual images having characteristics which are same as or similar to real feature maps derived from real images, by using GAN including a generating network and a discriminating network includes, in a learning device, steps of: allowing the generating network to apply convolutional operations to an input image, to thereby generate an output feature map having characteristics which are same as or similar to the real feature maps; and allowing a first loss unit to generate losses by referring to an evaluation score, corresponding to the output feature map, generated by the discriminating network.SELECTED DRAWING: Figure 2